Introducing the DeepSeek Harness Long-Term Memory Plugin
The `wwskills/dsh-long-memory` plugin addresses the lack of native long-term memory in DeepSeek Harness by enabling data persistence through SQLite storage and FTS5 full-text search. It features automatic memory extraction, a low-confidence confirmation queue, user profile construction, and self-evolving learning, allowing agents to optimize their behavior based on user corrections. The plugin provides a browser sidebar interface and a RESTful API, supports hybrid ranking with BM25, vector search, and RRF fusion, and is compatible with embedding models such as Ollama. Installation requires Node.js >= 22.5. Its flexible configuration makes it suitable for agent development involving multi-turn conversations and long-running tasks.
Read MoreDeepSeek Harness Long-Term Memory Plugin
The `dsh-memory-migration` plugin addresses the lack of native long-term memory in DeepSeek Harness. By leveraging local storage and the BM25 algorithm, it enables persistent key-value pair memory and intelligent recall, ensuring consistency in identity, preferences, and workflows across sessions. The plugin supports migrating memory from other platforms and provides a visual configuration interface. Data is stored locally, with a strong focus on privacy and security. Installation requires Node.js 22.19+ or 24+, making it suitable for scenarios that require maintaining a persona or reusing methodologies.
Read More